tracking with linear dynamic models t. darrell · 2002. 11. 22. · tracking and recursive...

6.801/866

Tracking with Linear Dynamic Models

T. Darrell

Tracking Applications

• Motion capture• Recognition from motion• Surveillance• Targeting

Things to consider in tracking

What are the• Real world dynamics• Approximate / assumed model• Observation / measurement process

This lecture focuses on models with linear dynamics and measurement process.

Outline

• Recursive filters• State abstraction• Density propagation• Linear Dynamic models• Kalman filter in 1-D• Kalman filter in n-D• Data association• Multiple models

(next: nonlinear models: EKF, Particle Filters)

Tracking and Recursive estimation

• Real-time / interactive imperative.• Task: At each time point, re-compute estimate of

position or pose.– At time n, fit model to data using time 0…n– At time n+1, fit model to data using time 0…n+1

• Repeat batch fit every time?

Recursive estimation

• Decompose estimation problem– part that depends on new observation– part that can be computed from previous history

• E.g., running average:at = α at-1 + (1-α) yt

Density propogation

• Tracking == Inference over time• Much simplification is possible with linear

dynamics and Gaussian probability models

Tracking

• Very general model: – We assume there are moving objects, which have an underlying

state X– There are measurements Y, some of which are functions of this

state– There is a clock

• at each tick, the state changes• at each tick, we get a new observation

• Examples– object is ball, state is 3D position+velocity, measurements are

stereo pairs– object is person, state is body configuration, measurements are

frames, clock is in camera (30 fps)

Three main issues in tracking

Simplifying Assumptions

Tracking as induction

• Assume data association is done– we’ll talk about this later; a dangerous assumption

• Do correction for the 0’th frame• Assume we have corrected estimate for i’th frame

– show we can do prediction for i+1, correction for i+1

Base case

Induction step

Given

Induction step

Linear dynamic models

• A linear dynamic model has the form

• This is much, much more general than it looks, and extremely powerful

xi = N Di−1xi−1;Σdi( )yi = N Mixi ;Σmi( )

xi = N Di−1xi−1;Σdi( )Observabilityyi = N Mixi ;Σmi( )

• For the measurement model, we may not need to observe the whole state of the object– e.g. a point moving in 3D, at the 3k’th tick we see x,

3k+1’th tick we see y, 3k+2’th tick we see z– in this case, we can still make decent estimates of all

three coordinates at each tick.

• This property is called Observability

xi = N Di−1xi−1;Σdi( )Examplesyi = N Mixi ;Σmi( )

• Random walk• Points moving with constant velocity• Points moving with constant acceleration• Periodic motion

xi = N Di−1xi−1;Σdi( )Examples

• Drifting points– assume that the new position of the point is the old one,

plus noiseD = Id

yi = N Mixi ;Σmi( )

cic.nist.gov/lipman/sciviz/images/random3.gif http://www.grunch.net/synergetics/images/random3.jpg

Constant velocity yi = N Mixi ;Σmi( )

xi = N Di−1xi−1;Σdi( )

• We have

– (the Greek letters denote noise terms)• Stack (u, v) into a single state vector

– which is the form we had above

ui = ui−1 + ∆tvi−1 + ε i

vi = vi−1 + ς i

uv

i

=1 ∆t0 1

uv

i−1

+ noise

xi = N Di−1xi−1;Σdi( )Constant acceleration

• We have

– (the Greek letters denote noise terms)• Stack (u, v) into a single state vector

– which is the form we had above

ui = ui−1 + ∆tvi−1 + ε i

vi = vi−1 + ∆tai−1 +ς i

ai = ai−1 + ξi

uva

i

=1 ∆t 00 1 ∆t0 0 1

uva

i−1

+ noise

yi = N Mixi ;Σmi( )

position

position

position andmeasurement

ConstantVelocityModel

velocity

time

time

time

position

position

velocity

ConstantAccelerationModel

xi = N Di−1xi−1;Σdi( )

Assume we have a point, moving on a line with a periodic movement defined with a differential eq:

can be defined as

with state defined as stacked position and velocity u=(p, v)

yi = N Mixi ;Σmi( )Periodic motion

xi = N Di−1xi−1;Σdi( )Periodic motionyi = N Mixi ;Σmi( )

Take discrete approximation….(e.g., forward Euler integration with ∆t stepsize.)

Higher order models

• Independence assumption

• Velocity and/or acceleration augmented position• Constant velocity model equivalent to

– velocity ==– acceleration ==– could also use , etc.

The Kalman Filter

• Key ideas: – Linear models interact uniquely well with Gaussian

noise - make the prior Gaussian, everything else Gaussian and the calculations are easy

– Gaussians are really easy to represent --- once you know the mean and covariance, you’re done

Recall the three main issues in tracking

(Ignore data association for now)

The Kalman Filter

[figure from http://www.cs.unc.edu/~welch/kalman/kalmanIntro.html]

The Kalman Filter in 1D

• Dynamic Model

• Notation

Predicted mean

Corrected mean

The Kalman Filter

Prediction for 1D Kalman filter

• The new state is obtained by– multiplying old state by known constant– adding zero-mean noise

• Therefore, predicted mean for new state is– constant times mean for old state

• Old variance is normal random variable– variance is multiplied by square of constant– and variance of noise is added.

The Kalman Filter

?…

Correction for 1D Kalman filter

Correction for 1D Kalman filter

Notice:– if measurement noise is small, we rely mainly on the measurement,– if it’s large, mainly on the prediction– σ does not depend on y

position

position

position andmeasurement

ConstantVelocityModel

velocity

time

time

time

position

position

velocity

ConstantAccelerationModel

Smoothing

• Idea– We don’t have the best estimate of state - what about

the future?– Run two filters, one moving forward, the other

backward in time.– Now combine state estimates

• The crucial point here is that we can obtain a smoothed estimate by viewing the backward filter’s prediction as yet another measurement for the forward filter

– so we’ve already done the equations

n-D

Generalization to n-D is straightforward but more complex.

n-D

Generalization to n-D is straightforward but more complex.

n-D Prediction

Generalization to n-D is straightforward but more complex.

Prediction:• Multiply estimate at prior time with forward model:

• Propagate covariance through model and add new noise:

n-D Correction

Generalization to n-D is straightforward but more complex.

Correction:• Update a priori estimate with measurement to form a

posteriori

n-D correction

Find linear filter on innovations

which minimizes a posteriori error covariance:

K is the Kalman Gain matrix. A solution is

( ) ( ) −− ++ xxxxE

T

Kalman Gain Matrix

As measurement becomes more reliable, K weights residual more heavily,

As prior covariance approaches 0, measurements are ignored:

1

0lim −

→Σ= MKi

m

0lim0

=→Σ−

iKi

2-D constant velocity example from Kevin Murphy’s Matlab toolbox• MSE of filtered estimate is 4.9; of smoothed estimate. 3.2. • Not only is the smoothed estimate better, but we know that it is better,

as illustrated by the smaller uncertainty ellipses• Note how the smoothed ellipses are larger at the ends, because these

points have seen less data. • Also, note how rapidly the filtered ellipses reach their steady-state

(“Ricatti”) values. [figure from http://www.ai.mit.edu/~murphyk/Software/Kalman/kalman.html]

Data Association

In real world yi have clutter as well as data…

E.g., match radar returns to set of aircraft trajectories.

Data Association

Approaches:• Nearest neighbours

– choose the measurement with highest probability given predicted state

– popular, but can lead to catastrophe

• Probabilistic Data Association– combine measurements, weighting by probability given

predicted state– gate using predicted state

Online demo

[figure from Welsh and Bishop 2001]

Online demo

The Kalman Filter Learning Tool tool simulates a relatively simple example setup involving estimation of the water level in a tank.

Water dynamics. The user can independently choose both the actual and modeled dynamics of the water. The choices include no motion (the default), filling, sloshing, or both filling and sloshing.

Measurement model. The user can also choose the method of measurement. The measurement model choices include two options that are commonly used (for example) in toilet tanks: a vertical level (linear) float-type sensor, or an angular (non-linear) float-type sensor. A diagram depicting the two case is shown below. The user is also allowed to increase or decrease (by a factor of 10) the magnitude of the random linear or angular measurement noise.

Online demo

[figure from http://www.cs.unc.edu/~welch/kalman/kftool/index.html]

Online demo

http://www.cs.unc.edu/~welch/kalman/kftool/KalmanFilterApplet.html

Abrupt changes

What if environment is sometimes unpredictable?

Do people move with constant velocity?

Test several models of assumed dynamics, use the best.

Multiple model filters

Test several models of assumed dynamics

[figure from Welsh and Bishop 2001]

MM estimate

[figure from Welsh and Bishop 2001]

P likelihood

[figure from Welsh and Bishop 2001]

No lag

[figure from Welsh and Bishop 2001]

Smooth when still

[figure from Welsh and Bishop 2001]

Resources

• Kalman filter homepagehttp://www.cs.unc.edu/~welch/kalman/

• Kevin Murphy’s Matlab toolbox:http://www.ai.mit.edu/~murphyk/Software/Kalman/k

alman.html

Outline

• Recursive filters• State abstraction• Density propagation• Linear Dynamic models• Kalman filter in 1-D• Kalman filter in n-D• Data association• Multiple models

(next: nonlinear models: EKF, Particle Filters)[Figures from F&P unless otherwise attributed]